Desertification risk assessment in arid regions is frequently hindered by the coarse spatial resolution of available soil datasets. Conventional linear approaches cannot fully meet the complex dynamic and static driving ecosystem in highly fragmented habitats. Consequently, it is often required to accurately identify the risk sources and micro-geomorphic variations in sustainable ecology. In this study, a high-precision desertification risk assessment was constructed in the extreme arid zones, thereby revealing the spatiotemporal evolution patterns of the ecosystem. A typical ecologically fragile region, the Tarim Populus Euphratica National Nature Reserve was selected as the study area. A framework was then developed as follows. Initially, a machine learning model with the random forest (RF) was employed to spatially downscale soil texture data from a 1 km resolution to 30 m. High-resolution remote sensing predictors were integrated with empirical field sampling data to enhance baseline accuracy. Subsequently, the comprehensive desertification risk level (CDRL) was formulated using a hierarchical zoning matrix. Dynamic wind erosion driving forces were also coupled to quantify by the revised wind erosion equation (RWEQ). The static ecosystem sensitivity was characterized by the desertification sensitivity index (DSI). Furthermore, the Theil-Sen Median trend analysis, the Mann-Kendall test, alongside the optimal multivariable stratified geodetector (OMGD) were applied to quantitatively decode the spatiotemporal evolution and the multi-factor nonlinear driving mechanisms of regional desertification risks over the period from 2000 to 2024. The results showed that: 1) The robust performance of spatial downscaling was achieved in the R² values from 0.79 to 0.82, a Kling-Gupta efficiency (KGE) between 0.64 and 0.70, after in-situ measurements. The high-resolution mapping also captured the spatial heterogeneity of micro-geomorphic features that were previously obscured. 2) The desertification risk in the reserve presents a gradient distribution pattern, being low along the river and high in peripheral areas. Extremely high-risk zones are mainly concentrated on the south bank of the Tarim River. 3) Temporal trend analyses revealed that the overall ecological status of the reserve maintained a resilient baseline. Furthermore, 73.64% of the area exhibited no significant changes over the study period. A spatial divergence emerged in the evolving areas: 12.82% of the area on the north bank shared the significant reduction of the risks, while 13.54% of the area on the southern edge was ongoing degradation, indicating a delicate dynamic balance. 4) The driving force detection indicated that the soil properties functioned as the dominant baseline factors, with their explanatory power (q values) consistently exceeding 0.91. Notably, the explanatory power of the vegetation quality index increased significantly from 0.69 in 2000 to 0.82 in 2024, indicating the growing positive impact of anthropogenic ecological water conveyance projects. In conclusion, the evaluation framework successfully facilitated a fine-scale, mechanism diagnosis of desertification risk, compared with the conventional coarse-scale ones. There was a dynamic balance between ecological restoration and natural degradation. According to these mechanistic insights, the differentiated spatial strategy was strongly recommended to implement targeted engineering for the 13.54% of degrading areas on the south bank. The findings can provide a strong reference for the ecological connectivity of the river corridor.

The main purpose of this research is to evaluate the effectiveness and practical performance of the spatiotemporal and temporal weighted regression model (GTWR), in order to accurately estimate the ozone (O₃) concentrations across China. A GTWR model was developed and validated to estimate O₃ concentrations, according to a comprehensive set of nine auxiliary variables. These variables included the ground-based O₃ monitoring station, Ozone Monitoring Instrument (OMI) ozone column concentration, relative humidity, precipitation, wind speed, temperature, evapotranspiration, atmospheric boundary layer height, normalized vegetation index (NDVI), and population density. The spatial distribution patterns of O₃ concentrations were analyzed using geographical detectors. A systematic investigation was then made to explore the influence of these nine driving factors on O₃ levels, particularly for the impact of these interactions among these factors on the governing mechanisms of O₃ distribution. The results reveal the following key points: 1) Multicollinearity and Model Performance: The nine variables shared a low level of multicollinearity, indicating the reliable performance of the model. GTWR model was achieved in a high level of accuracy over the period from 2014 to 2021, with the coefficient of determination (R²) not less than 0.81. The performance of the model also included a root mean square error (RMSE) ranging from 9.19 to 10.90 µg/m³ and a mean absolute error (MAE) between 6.27 and 7.73 µg/m³, indicating the robust predictive capabilities. 2) Trends in ozone concentrations: the annual average concentration of ozone demonstrated a general upward trend since 2014, characterized by an initial increase, a subsequent decrease, and a gradual rise. There were distinct seasonal variations in the O₃ levels. The average concentrations were ranked in the descending order of summer, spring, autumn, winter. The higher O₃ concentrations were observed during warmer seasons, compared with the cooler ones. The spatial distribution of O₃ concentrations shared a significant regional pattern that aligned closely with the latitude. This distribution of pattern also represented the population density and economic development across China, where the higher O₃ levels were concentrated in regions between 30° and 45° north latitude. 3) Geographical detector analysis: Evapotranspiration, atmospheric boundary layer height, and temperature were the strongest single factors on the O₃ level, with explanatory powers of 0.840, 0.797, and 0.759, respectively. The interactions of most factors shared a dual-factor enhancement, followed by a nonlinear enhancement, indicating joint changes in O₃. There was no interaction to show the linear or nonlinear weakening. All factors shared an enhancing effect on O₃ concentration, albeit to varying degrees. Furthermore, the explanatory power was further improved, when the factors interacted. Among them, the strongest interactions were observed between evapotranspiration and population density, as well as the relative humidity and temperature, with explanatory values of 0.95. Therefore, there was a more pronounced impact of factor interactions on O₃ concentrations, compared with single factors. Ecological detection showed significant differences between evapotranspiration, temperature, and all other factors except atmospheric boundary layer height, and between atmospheric boundary layer height and all other factors except evapotranspiration and temperature. It infers that the combined effects of evapotranspiration, atmospheric boundary layer height, and temperature with other factors posed a greater impact on the spatial distribution of ozone. Single-factor analysis also verified that these three factors shared a stronger effect on ozone. There was no significant difference between the rest factors, indicating their relatively similar mechanisms of impact on ozone concentration. In summary, the GTWR model was a robust tool to analyze the O₃ concentrations, and effectively capture both spatial and temporal variations. The findings also emphasized the complex interplay between environmental variables and O3 levels. The comprehensive models were necessary to consider both the individual and interactive effects of multiple factors. This approach can provide valuable insights into the spatial distribution and temporal dynamics of O₃.